Skip to content

Claude Code Without Coding · Section 10

Agents and Subagents, Complete Resource Pack

Every prompt, link and practice file from Section 10: give an agent a useful job, inspect what it receives and returns, and judge the result.

Arun Nagarathanam · 12 Oct 2026 Guide · 65 of 80 in this course

Lecture 10.1

What Does a Subagent Do?

See what a subagent is: a separate agent that works on one part of a larger job and brings its findings back.

Ideas to keep

  1. 01The main agent directs the overall work, supplies the assignment and uses the return.
  2. 02A subagent works on the part it received, in its own working conversation.
  3. 03Claude Code already has built-in agents. You can also save a named agent when a responsibility keeps returning.
  4. 04Willow Desk is an illustrative business with supplied practice files.

Follow along

  1. Download and extract the Section 10 practice ZIP.
  2. Open the extracted folder as a project in Claude Code.
  3. Read START-HERE.md before the first demonstration.

Files and links

section-10-practice-pack.zipSTART-HERE.md

Lecture 10.2

Give an Agent One Useful Job

Ask a built-in general-purpose subagent to critique a plan from a first-time customer's perspective.

Ideas to keep

  1. 01A plan can sound sensible while leaving a useful question unanswered.
  2. 02A perspective and a clear job focus the review: examine the plan, connect concerns to the comments and return questions.
  3. 03Check the comments the agent cites before treating its reading as evidence.
  4. 04The business decision stays with you.

Follow along

  1. Open business-plan.md and customer-feedback.md in the practice project.
  2. Send the critique prompt and open the running task to inspect the assignment.
  3. Open initial-plan-critique.md and check each concern against the feedback IDs.
Claude Code · Critique the plan with a subagent 
Use a general-purpose subagent to read demonstration-inputs/business-plan.md and demonstration-inputs/customer-feedback.md completely. Critique the proposed colour pilot from a first-time customer's perspective. Return up to 5 concerns or questions, each tied to a plan passage or feedback ID, with a useful next question to investigate. Leave both source files unchanged and keep the business decision with me. In the main conversation, save the returned critique unchanged as initial-plan-critique.md, read the report back and tell me which agent ran.
Paste into: Claude Code, in the extracted Section 10 practice folder

Files and links

demonstration-inputs/business-plan.mddemonstration-inputs/customer-feedback.md

Lecture 10.3

What Should Your Agent Receive and Return?

Inspect the assignment Claude actually sent to the subagent and what came back.

Ideas to keep

  1. 01A subagent starts with its own instructions and the task Claude provides, not your earlier conversation history.
  2. 02Separate context does not mean a separate computer: the files stay in the same project.
  3. 03A useful handoff names the purpose, the starting information, the actions allowed and the result needed back.
  4. 04The returned report still takes space in the main conversation, so ask for a concise return and keep detail in a named file.

Follow along

  1. Open the completed agent's task from the plan review.
  2. Compare the delegated prompt with the request you typed.
  3. Check that the purpose, files, perspective and limits survived the rewrite.

Files and links

Lecture 10.4

Create an Agent You Can Reuse

Save the plan review as a named project agent, customer-plan-critic, and call it by name.

Ideas to keep

  1. 01A custom agent is a Markdown file: opening fields identify it and set its configuration, and the body explains the work.
  2. 02The description tells Claude when the agent is useful; the name is what you use in a prompt.
  3. 03Read is enough when you name the files to examine.
  4. 04A newly created agents folder may need a new conversation before Claude can find the agent.

Follow along

  1. Ask Claude to create customer-plan-critic, checking the current documentation through Context7.
  2. Open .claude/agents/customer-plan-critic.md and read its name, description, tools, model and instructions.
  3. Call the agent by name; if it is not found, start a new conversation in the same project and try again.
  4. Answer the first question in knowledge-checks.md.
Claude Code · Create the critic agent 
Create a project agent called customer-plan-critic in this project. Check the current agent documentation through Context7 first. It reviews a supplied business plan and customer feedback from a first-time customer's perspective. Return up to 5 concerns, each supported by a plan passage or feedback IDs, with a useful next question to investigate. Put the concerns in an easy-to-read table and list missing information afterwards. Preserve the source files and keep business decisions with me.
Paste into: Claude Code, in the extracted Section 10 practice folder
Claude Code · Call the critic by name 
Use customer-plan-critic on demonstration-inputs/business-plan.md and demonstration-inputs/customer-feedback.md.
Paste into: Claude Code, in the extracted Section 10 practice folder

Files and links

knowledge-checks.md

Lecture 10.5

Can a Skill Call Agents?

Run the content-review Skill, which hands 3 different review jobs to general-purpose agents.

Ideas to keep

  1. 01A Skill can hold role instructions; Claude gives each role to a built-in general-purpose agent.
  2. 02The audience agent reads the 8 viewer comments, the performance agent compares the figures and the script agent reviews the written plan.
  3. 03The main agent checks the reports, updates the shared checklist and combines the findings.
  4. 04Save a Skill to reuse instructions; save a named agent when the same responsibility keeps returning.

Follow along

  1. Install content-review into this project's .claude/skills/ folder.
  2. Open the 3 inputs in demonstration-inputs/content-review/.
  3. Invoke the Skill, accept a review subfolder and open an agent's details while it runs.
  4. Read revised-video-brief.md when the main agent finishes.
Claude Code · Install the course Skills 
Install the Skills from skills/agent-fit and skills/content-review into this project's .claude/skills/ folder, preserving their complete folders and references. Read existing versions before replacing them and ask me if they differ. Show me the saved paths, read the installed files back and check that Claude can invoke each Skill. Install only these course Skills.
Paste into: Claude Code, in the extracted Section 10 practice folder
Claude Code · Run the content review 
/content-review Review demonstration-inputs/content-review/viewer-comments.csv, performance.csv and video-script.md in that same folder. Decide what the next version should explain or show more clearly, keeping a 45-second duration. Use separate audience, performance and script agents with distinct assignments. Save the three reports and revised-video-brief.md in review-output/. Preserve the inputs. The main conversation owns review-output/progress.md and tells me what finished.
Paste into: Claude Code, in the extracted Section 10 practice folder

Files and links

skills/content-review/SKILL.mddemonstration-inputs/content-review/video-script.mddemonstration-inputs/content-review/viewer-comments.csvdemonstration-inputs/content-review/performance.csvdemonstration-inputs/content-review/progress.md

Lecture 10.6

Give Different Agents Different Jobs

See why the 3 reviews could start together and why the combined brief had to wait for them.

Ideas to keep

  1. 01Agents can run in parallel when each has the input needed to answer its own question.
  2. 02A step that needs another agent's result waits until that result arrives.
  3. 03The main agent surfaced a disagreement (comment V05 against the script agent's timing) instead of hiding it.
  4. 04Watch times show something changed; the comments and script explain what to clarify.

Follow along

  1. Open the 3 completed agents' details and note what each received.
  2. Open revised-video-brief.md and check a recommendation against its quoted comment.
  3. Read the opening-order choice and decide which revision you would make.

Lecture 10.7

Research a Project Before Changing It

Use a research agent to investigate a local product page, then choose and approve a small change.

Ideas to keep

  1. 01A research agent reads and searches in its own conversation; the main conversation gets the findings.
  2. 02Research supplies reasons; you still choose the change.
  3. 03Describe each failed click so Claude can repair the right thing.
  4. 04Ask Claude to show suggestions and wait for approval before additional content changes.

Follow along

  1. Open product-page.html and setup-guide.html and try their links.
  2. Send the research prompt and open setup-research-brief.md.
  3. Choose one change and send the editing prompt.
  4. Open the page in the browser, click the guide link and report anything that fails.
  5. Answer the second question in knowledge-checks.md.
Claude Code · Research before changing 
Use a research agent to read demonstration-inputs/product-page.html, demonstration-inputs/setup-guide.html and demonstration-inputs/customer-feedback.md. Identify unanswered setup questions, use web searches where relevant and suggest 3 useful improvements. Each suggestion should explain which feedback it addresses. Save the findings as setup-research-brief.md and leave the original files unchanged.
Paste into: Claude Code, in the extracted Section 10 practice folder
Claude Code · Make the chosen edit 
Use any design Skill that you have and add a short setup section to the product page and a clear link to our existing setup guide. Use the guide's actual information.
Paste into: Claude Code, in the extracted Section 10 practice folder
Claude Code · Repair and propose, then wait 
The full-guide link on the product page does not open the guide, the product-details link does not work, and the guide's return link does not open the product page. Repair those links. Use the same design Skill to improve both pages. For any additional content changes, show me your suggestions and wait for my approval.
Paste into: Claude Code, in the extracted Section 10 practice folder

Files and links

demonstration-inputs/product-page.htmldemonstration-inputs/setup-guide.html

Lecture 10.8

Choose What an Agent Can Do

Match an agent's tools and model to its job, and keep your approval for real actions.

Ideas to keep

  1. 01The tools field decides what an agent can do; the instructions decide which file and what content.
  2. 02A saved model and effort choice applies each time the agent runs.
  3. 03Every agent's usage counts alongside the main conversation.
  4. 04maxTurns limits rounds of work; it is not a cost limit or a schedule.
  5. 05Agent teams coordinate directly with one another; subagents return to the main conversation.

Follow along

  1. Open customer-plan-critic.md and read its tools and model fields.
  2. Ask the critic to save its own report and approve adding Write.
  3. Start a fresh conversation to try the updated agent, then open the report.
  4. Read the optional maxTurns example in AGENT-CONFIGURATION.md.
Claude Code · Let the critic save its own report 
Use customer-plan-critic to review the plan and feedback again, and have the agent itself save its report as customer-plan-critique.md in this project.
Paste into: Claude Code, in the extracted Section 10 practice folder
Claude Code · Approve the tool change 
Yes, add Write and use that filename.
Paste into: Claude Code, in the extracted Section 10 practice folder

Files and links

AGENT-CONFIGURATION.md

Lecture 10.9

Find and Adapt Existing Agents

Adapt a community agent (VoltAgent's content-marketer) into a project agent for one setup-post job.

Ideas to keep

  1. 01An existing agent is a starting point to inspect, not a finished fit.
  2. 02Keep the useful instructions and remove work the project does not need.
  3. 03Keep the source attribution and MIT licence beside the adapted file.
  4. 04Compare the draft with the source guide before changing or publishing anything.

Follow along

  1. Open the VoltAgent Business and Product collection and read content-marketer.
  2. Ask Claude to adapt it into setup-content-writer for the Willow Desk guide.
  3. Open the saved definition, then run it in a new conversation if it is not found.
  4. Compare the draft with setup-guide.html and ask for specific revisions.
Claude Code · Adapt an existing agent 
Use VoltAgent's content-marketer agent as a starting point: [paste the content-marketer link from this lecture's website section].
Adapt it for Willow Desk. It should read our setup guide and draft one helpful LinkedIn post in a simple conversational tone. Keep the setup information accurate and return the draft for me to review. Remove instructions for campaigns or services this project does not use. Save it as a project agent called setup-content-writer.
Paste into: Claude Code, in the extracted Section 10 practice folder
Claude Code · Run the adapted agent 
Use setup-content-writer to turn our setup guide into a helpful LinkedIn post. Return a draft for me to review.
Paste into: Claude Code, in the extracted Section 10 practice folder

Files and links

EXISTING-AGENTS.mddemonstration-inputs/agent-discovery/

Lecture 10.10

Which Job Deserves an Agent?

Use the agent-fit Skill to decide whether a job deserves an agent, then work through the 5 common problems.

Ideas to keep

  1. 01agent-fit asks one question at a time: up to 5, then up to 5 more if an important uncertainty remains.
  2. 02A recommendation can be: keep it in the main conversation, give an existing agent one part, create a reusable agent or refine the idea.
  3. 03A saved definition and a successful run are separate checks.
  4. 04The 5 problems: choosing the work, a misunderstood job, the wrong agent, agents without a useful split, and not knowing what finished.

Follow along

  1. Install agent-fit (AGENT-FIT-INSTALL.md) or use AGENT-FIT-PROMPT.md in the conversation.
  2. Describe a recurring job and answer the questions.
  3. Read the proposed brief and refine any rule that is broader than the job needs.
  4. Open FIVE-PRACTICAL-SOLUTIONS.md and pick the prompt for your actual problem.
  5. Answer the third question in knowledge-checks.md.
Claude Code · Ask agent-fit about a recurring review 
I want an agent to review our business plan and customer feedback every week. Help me decide whether that's useful. I want to keep the final business decision with me.
Paste into: Claude Code, in the extracted Section 10 practice folder
Claude Code · Confirm creation 
Yes, create it.
Paste into: Claude Code, in the extracted Section 10 practice folder

Files and links

skills/agent-fit/AGENT-FIT-INSTALL.mdAGENT-FIT-PROMPT.mdFIVE-PRACTICAL-SOLUTIONS.mdAGENT-REPAIR-PROMPT.mdAGENT-COMPLETION-PROMPT.mdAGENT-FEEDBACK-AND-PROGRESS.md

Lecture 10.11

Choose Your Next Delegation

Choose one job of your own to delegate, with a result you can inspect.

Ideas to keep

  1. 01Pick the part of the work an agent can take, and what you need back to use its return.
  2. 02For a review, name the questions and the material; for a change, name the target and the action you authorize.
  3. 03Independent reviews can run together; an edit chosen from research waits for the research.
  4. 04If the first return misses something, explain what you expected and why.

Follow along

  1. Choose a plan, source material or project that matters to you.
  2. Use agent-fit, or follow the demonstrated route with the practice files.
  3. Inspect the return and improve the instructions for the next run.
Claude Code · Start your own delegation 
/agent-fit I want to improve [describe your work and desired result]. Help me decide whether an agent would be useful. Ask one question at a time and refine the next question from my answer.
Paste into: Claude Code, in the extracted Section 10 practice folder

Practice files

More in this section

Files and links

Keep building

More from Aruntastic

Practical courses for people who want to build useful things with AI.

AI SEO Course 2026: From Zero to Page-One Results

Research search demand, structure useful pages, and work through technical and off-page SEO.

Open the course →

AEO, GEO, AI SEO: Get Cited by ChatGPT, Google, Perplexity

Study how AI search finds information, and make your pages clear enough to be cited.

Open the course →

Microsoft PowerPoint Presentation 2026 for Professionals

Learn visual hierarchy, slide composition, typography and animation through practical work.

Open the course →

AI Agents for Leaders Who Have No Time to Learn AI

If you can delegate to a person, you can delegate to an AI agent. You keep the judgment. For managers and founders.

Open the course →

Introduction to Agentic AI and AI Agents

Learn how to direct an AI agent at work, with agentic workflows, RAG and MCP explained from a real first run.

Open the course →

Toolkits and programme

Go deeper

Toolkit

SEO Accelerator Toolkit

150+ copy-paste AI prompts for audience research, keywords, content, technical SEO and link building.

Toolkit

GEO Accelerator Toolkit

100 detailed AI prompts for GEO audits, entity building, content structure and client reporting.

Programme

GEO Authority Accelerator

A complete GEO programme on our learner platform: nine modules and a 30, 60, 90 day roadmap. The GEO Accelerator Toolkit is included.

Claude certification practice

Preparing for a Claude certification?

Practice exams written by Arun: 600 questions across 4 mock exams and 2 drill banks. Unofficial preparation, not affiliated with Anthropic.

CCAR-FArchitect FoundationsCCAO-FAssociate FoundationsCCDV-FDeveloper FoundationsCCAR-PArchitect Professional

See Where You Stand

This resource is one piece of a much bigger system.

Try the free Claude Code Readiness Quiz alongside the course.

Take the Claude Code Readiness Quiz Free course readiness quiz